CreditCard Fraud Detection Support Vector Machines¶
Data Description¶
There are 25 variables:
- ID: ID of each client
- LIMIT_BAL: Amount of given credit in NT dollars (includes individual and family/supplementary credit
- SEX: Gender (1=male, 2=female)
- EDUCATION: (1=graduate school, 2=university, 3=high school, 4=others, 5=unknown, 6=unknown)
- MARRIAGE: Marital status (1=married, 2=single, 3=others)
- AGE: Age in years
- PAY_0: Repayment status in September, 2005 (-1=pay duly, 1=payment delay for one month, 2=payment delay for two months, ... 8=payment delay for eight months, 9=payment delay for nine months and above)
- PAY_2: Repayment status in August, 2005 (scale same as above)
- PAY_3: Repayment status in July, 2005 (scale same as above)
- PAY_4: Repayment status in June, 2005 (scale same as above)
- PAY_5: Repayment status in May, 2005 (scale same as above)
- PAY_6: Repayment status in April, 2005 (scale same as above)
- BILL_AMT1: Amount of bill statement in September, 2005 (NT dollar)
- BILL_AMT2: Amount of bill statement in August, 2005 (NT dollar)
- BILL_AMT3: Amount of bill statement in July, 2005 (NT dollar)
- BILL_AMT4: Amount of bill statement in June, 2005 (NT dollar)
- BILL_AMT5: Amount of bill statement in May, 2005 (NT dollar)
- BILL_AMT6: Amount of bill statement in April, 2005 (NT dollar)
- PAY_AMT1: Amount of previous payment in September, 2005 (NT dollar)
- PAY_AMT2: Amount of previous payment in August, 2005 (NT dollar)
- PAY_AMT3: Amount of previous payment in July, 2005 (NT dollar)
- PAY_AMT4: Amount of previous payment in June, 2005 (NT dollar)
- PAY_AMT5: Amount of previous payment in May, 2005 (NT dollar)
- PAY_AMT6: Amount of previous payment in April, 2005 (NT dollar)
- default.payment.next.month: Default payment (1=yes, 0=no)
In [1]:
import numpy as np # linear algebra
import pandas as pd # data processing, CSV file I/O (e.g. pd.read_csv)
import seaborn as sns
%matplotlib inline
import matplotlib.pyplot as plt
import matplotlib.colors as colors
from sklearn.utils import resample
from sklearn.preprocessing import scale
from sklearn.decomposition import PCA
from sklearn.model_selection import train_test_split
from sklearn.model_selection import GridSearchCV, StratifiedKFold
from sklearn.svm import SVC
from sklearn.metrics import classification_report, confusion_matrix
from sklearn.metrics import accuracy_score, roc_auc_score, plot_confusion_matrix, plot_roc_curve, plot_precision_recall_curve
import os
for dirname, _, filenames in os.walk('/kaggle/input'):
for filename in filenames:
print(os.path.join(dirname, filename))
In [2]:
file_path = "/kaggle/input/default-of-credit-card-clients-dataset/UCI_Credit_Card.csv"
#file_path1 = "https://archive.ics.uci.edu/ml/machine-learning-databases/00350/default%20of%20credit%20card%20clients.xls"
df = pd.read_csv(file_path)
df.head()
Out[2]:
In [3]:
df.info()
In [4]:
df.rename({"default.payment.next.month" : "DEFAULT"}, axis="columns", inplace=True)
df.head()
Out[4]:
In [5]:
df.drop(columns='ID', inplace=True)
df.head(3)
Out[5]:
Statistical Five Number Summary¶
In [6]:
df.describe().T
Out[6]:
Identifying Missing Data¶
In [7]:
df.info()
In [8]:
print("Unique values of each column\n")
for cols in ['SEX', 'EDUCATION', 'MARRIAGE', 'AGE', 'DEFAULT']:
print(cols, " : ", df[cols].unique())
In [9]:
len(df.loc[(df['EDUCATION']==0) | (df['MARRIAGE']==0)])
Out[9]:
In [10]:
len(df)
Out[10]:
In [11]:
len(df.loc[(df['EDUCATION']==0) | (df['MARRIAGE']==0)]) / len(df) * 100
Out[11]:
Percentage of missing values in 0.23%. Hence, we can remove them from the analysis.
In [12]:
df_msno = df.loc[(df['EDUCATION']==0) | (df['MARRIAGE']==0)]
df_msno.shape
Out[12]:
In [13]:
df = df.loc[(df['EDUCATION'] != 0) & (df['MARRIAGE'] != 0)]
df.shape
Out[13]:
In [14]:
df['DEFAULT'].value_counts()
Out[14]:
In [15]:
df_default = df[df['DEFAULT']==1]
df_no_default = df[df['DEFAULT']==0]
display(len(df_default), len(df_no_default))
In [16]:
sns.countplot(df['SEX']);
In [17]:
sns.countplot(df['MARRIAGE']);
In [18]:
sns.countplot(df['EDUCATION']);
In [19]:
sns.distplot(df['LIMIT_BAL']);
In [20]:
default_by_gender = pd.crosstab(df['SEX'], df['DEFAULT'])
sns.heatmap(default_by_gender, annot=True, fmt='2d');
In [21]:
default_by_gender.plot(kind='barh', stacked=True);
Pairwise Plots of the variables¶
In [22]:
plt.figure(figsize=(18,18))
sns.pairplot(df)
plt.show()
In [23]:
plt.figure(figsize=(15,15))
sns.heatmap(df.corr(), annot=True, fmt='.2f', square=True)
plt.show()
Resampling¶
In [24]:
df_no_default_downsampled = resample(df_no_default,
replace=False,
n_samples=1000,
random_state=24)
len(df_no_default_downsampled)
Out[24]:
In [25]:
df_default_downsampled = resample(df_default,
replace=False,
n_samples=1000,
random_state=24)
len(df_default_downsampled)
Out[25]:
In [26]:
df_downsample = pd.concat([df_no_default_downsampled, df_default_downsampled])
len(df_downsample)
Out[26]:
Separating the Independent and Dependent Variables¶
In [27]:
X = df_downsample.drop(columns='DEFAULT', axis=1).copy()
X.shape
Out[27]:
In [28]:
y = df_downsample['DEFAULT'].copy()
y.shape
Out[28]:
One-Hot Encoding for Categorical Variables¶
In [29]:
X_encoded = pd.get_dummies(X, columns=['SEX','MARRIAGE','EDUCATION','PAY_0','PAY_2','PAY_3','PAY_4','PAY_5','PAY_6'])
X_encoded.head()
Out[29]:
In [30]:
X_encoded.shape
Out[30]:
Split into Train and Test sets¶
In [31]:
X_train, X_test, y_train, y_test = train_test_split(X_encoded, y, test_size=0.3, random_state=24)
In [32]:
X_train.head(3)
Out[32]:
In [33]:
X_test.head(3)
Out[33]:
Standard Scaling for using Radial Basis Function Kernel¶
In [34]:
X_train_scaled = scale(X_train)
X_test_scaled = scale(X_test)
Build SVC using RBF Kernel¶
In [35]:
clf_svc = SVC(C=1.0,
kernel='rbf',
gamma='auto',
probability=True)
In [36]:
clf_svc.fit(X_train_scaled, y_train)
Out[36]:
In [37]:
y_pred = clf_svc.predict(X_test_scaled)
In [38]:
print("Classification Report : \n")
print(classification_report(y_pred, y_test))
In [39]:
plot_confusion_matrix(clf_svc,
X_test_scaled,
y_test,
values_format='d',
display_labels=['No Default','Default'])
Out[39]:
In [40]:
plot_roc_curve(clf_svc,
X_test_scaled,
y_test)
Out[40]:
In [41]:
plot_precision_recall_curve(clf_svc,
X_test_scaled,
y_test)
Out[41]:
Hyperparameter Tuning with GridSearchCV¶
In [42]:
param_grid = [{
'C' : [0.5, 1.0, 10, 100],
'gamma' : ['scale', 1, 0.1, 0.01, 0.001, 0.0001],
'kernel' : ['rbf']
}]
In [43]:
clf_svc_tuned = GridSearchCV(SVC(),
param_grid,
cv=5,
scoring='accuracy',
verbose=2
)
In [44]:
clf_svc_tuned.fit(X_train_scaled, y_train)
Out[44]:
In [45]:
clf_svc_tuned.best_estimator_
Out[45]:
In [46]:
clf_svc_tuned.best_params_
Out[46]:
In [47]:
y_pred_tuned = clf_svc_tuned.predict(X_test_scaled)
In [48]:
print("Classification Report : \n")
print(classification_report(y_pred_tuned, y_test))
In [49]:
plot_confusion_matrix(clf_svc_tuned,
X_test_scaled,
y_test,
values_format='d',
display_labels=['No Default','Default'])
Out[49]:
In [50]:
plot_roc_curve(clf_svc_tuned,
X_test_scaled,
y_test)
Out[50]:
In [51]:
plot_precision_recall_curve(clf_svc_tuned,
X_test_scaled,
y_test)
Out[51]:
Plotting the Decision Boundary of SVC¶
In [52]:
pca = PCA()
In [53]:
X_train_pca = pca.fit_transform(X_train_scaled)
In [54]:
pvar = np.round(pca.explained_variance_ratio_*100, decimals=1)
labels = [str(x) for x in range(1, len(pvar)+1)]
plt.bar(x=range(1, len(pvar)+1), height=pvar)
plt.tick_params(axis='x',
which='both',
bottom=False,
top=False,
labelbottom=False)
plt.ylabel('Percentage of Explained Variance')
plt.xlabel('Principal Components')
plt.title('Scree Plot')
plt.show()
In [55]:
train_pc1_coords = X_train_pca[:,0]
train_pc2_coords = X_train_pca[:,1]
pca_train_scaled = np.column_stack((train_pc1_coords, train_pc2_coords))
In [56]:
clf_svc_tuned.fit(pca_train_scaled, y_train)
Out[56]:
In [57]:
X_test_pca = pca.transform(X_train_scaled)
In [58]:
test_pc1_coords = X_test_pca[:,0]
test_pc2_coords = X_test_pca[:,1]
pca_test_scaled = np.column_stack((train_pc1_coords, train_pc2_coords))
x_min = test_pc1_coords.min() - 1
x_max = test_pc1_coords.max() - 1
y_min = test_pc2_coords.min() - 1
y_max = test_pc2_coords.max() - 1
xx, yy = np.meshgrid(np.arange(start=x_min, stop=x_max, step=0.1),
np.arange(start=y_min, stop=y_max, step=0.1))
Z = clf_svc_tuned.predict(np.column_stack((xx.ravel(), yy.ravel())))
Z = Z.reshape(xx.shape)
fig, ax = plt.subplots(figsize=(10,10))
ax.contourf(xx, yy, Z, alpha=0.1)
cmap = colors.ListedColormap(['#e41a1c','#4daf4a'])
scatter = ax.scatter(test_pc1_coords,
test_pc2_coords,
c=y_train,
cmap=cmap,
s=100,
edgecolors='k',
alpha=0.7)
legend = ax.legend(scatter.legend_elements()[0],
scatter.legend_elements()[1],
loc='upper right')
legend.get_texts()[0].set_text('No Default')
legend.get_texts()[1].set_text('Default')
ax.set_ylabel('PC2')
ax.set_xlabel('PC1')
ax.set_title('Decision Boundary')
plt.show()
In [ ]: